Gadgetory

we've all seen the discussion as to how expensive HP m2 is but there haven't been many firm numbers attached to the perk card cost of Andy's decision to opt for the on interposer solution and the standalone MSRP structure for Vega 56 positions it competitively with the gtx 1070 it carries comparable performance equal memory capacity and similar target retail price assuming things calm down for the entire GPU market at some point given HB M 2's higher cost and Vega 56 is a bigger die that leaves little room for Andy the profit when compared to the gddr5 solutions that's what we're exploring today alongside y and you had to use HP m2 before getting to that this video is brought to you by synergy the software that lets you share a keyboard and mouse between multiple systems if you have limited desk space and multiple computers to command synergy removes the need for a separate peripherals or a KVM and works as over the network software use our link below to get 50% off the home or pro version with ssl there are a lot of reasons that Andy went with HP m2 of course it's not just a marketing thing of this is the new shiny object that we have you should buy it there are actual legitimate reasons especially on the power front and power consumption with Vega architecture we'll get to those in a little bit in this video but just the basics here for folks asking why Andy can't just make a bag of 56 or 64 with gddr5 you can't just swap between the two different types of memory it would require a new memory controller setup do a lot of things really a new power delivery system so it's not quite that easy they're really stuck with what they've got now and there are reasons that they went for HP m2 but it is factually more expensive than gddr5 so walk through both of those aspects today let's start with prices and then talk strategy and these pricing structure for Vega uniquely leans on bundle packs to help improve the company's value argument in a competitive market MSRP is $400 on rx bacon 56 500 on rx Bega 64 and then an added $100 up charge in exchange for two games and some instant discounts and these intention with this is to offer greater value to gamers but clearly this will also help the company increase margins we have more rise in parts and thus recoup some of the potential low or negative margins on Vega this is aided particularly with game bundles where a be partners pay Andy about $29 for the game codes though that is often waived or offered in exchange for MDF and the also stated desire to save off some mining purchases with the increased bundle prices as this would offset the value proposition of the card to minors since the bundles are sold as standalone skews and can't be broken by consumers ie broken into different pieces it seems that this is potentially an effective solution at keeping minors at bay if that were indeed the intention this moved to leverage bundle packs which is somewhat unique to both GP manufacturers is also mystified by complex economic surrounding potential long term plays by Andy with the company having already somewhat lost a bet on HP m2 pricing decreasing by the time Vega rolled out in fact price increased at least once within the past year and Hynix ultimately failed to deliver on Andy's demands in a timely fashion or a an achievable fashion at that in the meantime Andy could have its outlook set on increased supply driving down costs to build Vega GPUs this might mean taking a slight loss at the beginning or running on slim margins for now with the hopes of a payoff somewhere down the line part of this hinges on hynek's potentially coming online with HP on to at some point which would hypothetically reduce the cost figures that will go over in this video but ultimately Andy also needs to reclaim gaming market share part of that reclamation process will be aided by gritting teeth through painfully slim margins or even losses at launch and of course there's a reason that high necks couldn't meet demand for now part of it is that Andy is really their only large demand partner in HBM 2 so they didn't have a whole lot to deliver on other than Andy's contract there are two major costs with a video card the ASIC and the memory with follow up costs comprised of the VR MN to a lesser extent the cooler the PCB all the penny a dozen parts let's start with the HB m 2 and interpose reprising as that's what we're confident in speaking with David Cantor of real-world tech the analyst who broke news on Maxwell's tile-based rasterization and who previously worked at microprocessor report we received the following estimate from David Cantor quote the HBM to memory is probably around $150 and the interposer in packaging should be 25 we later compared this estimate with early rumors of HB m to pricing from earlier year and word from four different vendors he spoke with gamers Nexus independently all of which were within five to ten dollars of each other and with David Cantor's on record estimate this gave us high confidence in the numbers taking his $175 estimate of combined HP m2 and interpose her figure we're nearly halfway to the MSRP of Vega 56 with the rest of the cost comprised of the usual the VRM the GPU dye itself which we have no idea what that costs and diamond does and electrical components it cost a normal person for instance about forty five dollars to build the vrm on Vega that included the two dollar 70 cent per phase cost for instance of the irf 6890 fours and the IRF 68:11 low and high side direct FETs about eight dollars eighty cents for all six of the IR 3598 drivers and roughly four dollars on the IR 35 to 17 voltage controller Andy is a large company though and would receive volume discounts even as individuals we could order 10,000 of these parts and drive that cause down so we don't know what they pay for the vrm and we're not sure how many AMD ordered and aren't going to speculate on what their discount would be but those numbers give an idea for what someone might pay if they weren't a major corporation this primarily helps explain why and the opted for the same PCB and vrm on Vega F II 64 and 56 especially given the BIOS and power log on 56 although Vega can certainly benefit from the advanced vrm the necessity of it lessens as we get down to the lower end card the increased volume from v 56 orders could offset cost across the entire product stack to a point of it being cheaper to over build Vega 56 then to order two or three completely different sets of vrm components and PCBs regardless were about a hundred and fifty dollars in on the HBM two and twenty-five dollars on the inner poseur putting us at around 175 with the cost of the memory system before the die or any of the other components we did speak with numerous folks including cantor on estimated GPI costs and pricing on that for Andy and for the AV partners but the estimates were ranging massively I like hundreds of dollars of range to the point where they were really just educated guesses and without something more concrete to work with without corroboration between all the different estimates on record and off there's no point in trying to speculate on them so we do not know the cost of the GPU dye we're only focusing on the big thing here which is what everyone talks about HP m2 is expensive that's how expensive it is but the next part is how expensive is gddr5 because you need a point of comparison there to make any sense of it ultimately for GPU cost and yield only really AMD and global foundry know that for sure and so then the next question is what about the gddr5 costs a recent digitized report pegs gddr5 at about six dollars fifty cents for an eight gigabit module though also shows pricing for August onward at 850 per module with the old pricing let's run fifty two dollars cost for an eight gigabyte card or sixty eight dollars with the new pricing we do not presently know the g5x cost but the cost of gddr5 puts HBM to at about three times the cost for an equivalent amount without factoring and yields or the larger GPU die of course so this shows why and these margins are so thin on Vega we also know that Andy is passing along its HP m2 cost to partners at roughly a one-to-one rate they're not really up charging on it which is typically what happens with gddr5 memory there's no room to up charge the HBM to with vega's price target though ignoring GPU cost and cost of less significant components again like the vrm and cooler we're at 100 to $130 more than 8 gigabytes of gddr5 cost to build this is also ignoring other costs like incalculable R&D or packaging costs we're just focusing on memory today as that's what we know at the greatest accuracy now that we know how much HP m2 and gddr5 cost it's important to talk about why Andy's using HP m2 they're spending a hundred plus dollars more for the same capacity so there's got to be a reason for it beyond marketing and beyond what they normally say so Andy generally puts it this way like most of AMD's hardware the company is looking at HP m2 as a long-term investment they're looking at it as a long play technologically to try and get some kind of benefit out of it but the real reason the most important reason is that HBM to critically allows Andy to reduce its power consumption in terms of bandwidth per watt so to speak because gddr5 really wouldn't enable that with Vega architecture Vega architecture on its own draws clearly plenty of power but without HBM - it'd be a whole lot more and ultimately this is something we talked out with Fiji with the fury X is carrying through through Vega with an upgrade to the HBM now in iteration two and it's the same story it's that using HP m2 allows AMD to lower their power consumption and what their architecture that is particularly required it's a very specific need on a per Architecture basis we're presently Nvidia can get away without HB m2 on their consumer cards but NVIDIA is in a much different market position they have a much different architecture and AMD built for HB m and their GPU Vega is particularly starved for memory bandwidth without HB m2 or even with it if you run it at a lower frequency so there's a clear need for this memory design on Vega so back to the power side of the arguments begin with build Zoid we know that vega frontier Edition 16 gigabyte HBM 2 pulls 20 watts max using a DMM to determine this consumption the signor is the voltage controllers 3.3 volt draw but we're still at 20 watts on memory and no more than an additional 10 watts for the controller so that's less than 30 watch the entire memory system on Vega Frontier addition we also know that an R X 480 uses 40 to 50 watts for its eight gigabytes of memory which is already a significant increase in power consumption per gigabyte over Vega Fe again Fe the R X 480 also has a memory bandwidth of 256 gigabytes per second with 8 gigabytes of gddr5 versus Vega 56 is 484 gigabytes per second the result is increased bandwidth the same capacity and lower power consumption but at a much higher cost to build in order for an Rx for a to hypothetically reach similar bandwidth power consumption would increase significantly build joy calculates that a hypothetical 384-bit gddr5 bus on polaris architecture would push 60 to 75 watts and a 512 bit bus would do 80 to 100 watts for this reason alone HBM 2 saves AMD from a high power budget that would otherwise be spent solely on memory this comes down to architectural decisions made years ago by Andy at this point which are most readily solved for with HB m2 as HB m to provides greater bandwidth per watt than gddr5 HB m is effectively a necessity to make Vega at least somewhat power efficient while keeping the higher memory bandwidth imagine Vega 56 or 64 drawing an additional 80 to 100 watts the world wouldn't have it the world freaked out as it is at what Vega draws now and it'd be one of the hottest cards made since the GTX 480 or the r9 290x the Vega architecture is clearly starved by memory bandwidth to overclock an HB m - alone shows this as its gains are greater than just core overclock increases and they didn't have another choice but to go with HB m - this was forced even though cost would be roughly 1/3 on the other type of memory gddr5 it just didn't make sense to use it gddr5 might be possible but not without blowing power consumption through the roof or losing performance by limited bandwidth the alternative is just a lower performing card and they provided GN with a statement pertaining the choices revolving around HB m - which reads as follows am the chose a CH BM to memory for Vega because this advanced memory technology has clear benefits on multiple fronts Adri m2 is a second generation product that offers nearly twice the bandwidth per pen of first generation HB m thanks to various refinements as we noted in the vega white paper HB m to offers over three times the bandwidth per watt compared to gddr5 each stack of HB m2 has a wide dedicated 1024 bit interface and side note here there are two stacks of that allowing the memory devices to run at relatively low clock speeds while delivering tremendous bandwidth also thanks to die stacking and the use of an interpose err they go with HB m2 achieves a 75% smaller footprint for the GPU die plus memories versus a comparable gddr5 solution finally the combination of high bandwidth excellent power efficiency and a compact physical footprint made HBM - a clear choice for Vega we have no plans to step back to gddr5 and Andy did end up going for two stacks of HB m2 rather than four so compared to Fiji we're at roughly the same memory bandwidth as the previous version of HB m just with two stacks they're actually technically it's a slight decrease in memory bandwidth but the architecture is different overall so not quite that simple either way Andy says no plans to go back to gddr5 they do benefit from the bandwidth per watt category for sure and that's ultimately the crux of this decision more so than anything else more than any future looking marketing more than any of the usual items that you may have read pertaining to this launch it is lowering the power consumption for an equivalent amount of memory with ultimately a better bandwidth than they would have gotten with gddr5 that's why hbm's who is used as for cost knowing that the memory system gets us up to nearly $200 as a starting point it's in arguable that Andy has lower margins on Vega products than could be had with gddr5 but the company also didn't have a choice but to use HP m2 it's not a good or a bad thing it's just a fact Nvidia forced Andy's hand here they dropped the 1080 Ti in March alongside Rison therefore stealing some of the thunder from Rison pulling in more sales because you really didn't have another option but on video to pair with those 1,800 X CPUs and then they followed that with a 1070 and 1080 MSRP reduction that's ignoring the current insane GPU pricing and just looking at MSRP as that's ultimately where the two companies battle under normal conditions AMD is able to make back some of these margins with bundle packs where a pair of games can be sold to a IBS for $29 cents consumers for $100 despite the games presently at time of shooting being available for 90 they can also make back margin where risin and motherboard parts help recoup some of the additional spend each motherboard sold is and other chips that moved and rise in sales go completely to AMD either way Andy has to increase its GPU market share fighting through early losses or slim margins as part of that and the long-term play is clearly hoping that increased demand and supply will lower cost a bill so it remains to be seen how that will play out well there's a bundle thing certainly helps in the interim so that gives us a starting point for cost to helps contextualize what people mean when they say HP m2 is expensive that's your number that's how expensive it is it's about three times more than gddr5 give or take $20 depending on the gddr5 market right now and that's pretty rough for Andy's profitability and these making a lot of plays here to try and recoup some of they're doing the bundle packs for one and they're selling cards that are basically well they've got a couple different spins on the same PCB and BRM that helps a lot too because ordering more volume of those parts means you don't have to one order a special small quantity for Vega frontier edition and then a different quantity for Vega 64 and then a completely different board for Vega 56 so you save money there that's one of the other plays this one as a specific benefit to consumers means that if you buy bacon 56 you do actually get a really good vrm so that's a nice benefit but it is also to help them reduce cost additionally the company really couldn't have reasonably gotten gddr5 to work on Vega without major drawbacks based on what we know of the architecture based on you check the power consumption yourself of the memory at less than 30 watts for 16 gigabytes of HBM - you can't compete with that with gddr5 not with big it's to memory starved they'd have to go with a smaller bus they do so at a higher power consumption and it just wouldn't be a good look it wouldn't be a good product that could compete now it would be cheaper to make so you might end up with another Polaris where it's just kind of another mid-range product rather than targeting some high-end so Andy really was in a tough spot they were forced to do this HP m2 is not used because Andy is trying to be the good guy and further the market here and push technology it's being used out of necessity it would be nice if everyone started using HP MT because I mean ultimately it is objective fact that it's just more efficient in terms of performance or bandwidth per watt than gddr5 g5x is a different story I'm not talking about that don't know enough about its cost or anything like that HBM - certainly would benefit a 1070 for example but the difference is and video and AMD have different needs and Vidya doesn't need HBM - to get the performance it gets its architecture is different and frankly speaking in terms of market Nvidia doesn't feel the same pressure that AMD does Andy is pressured into making decisions that are more costly for them or more risky but that's what you have to do when you try to compete where you're 30% to 70% or whatever it is these days so that's where it stands there's there's your pricing anyone who is curious there's why they used it and as far as performance we're doing more work with Vega 56 hybrid the mod that we live streamed recently overclocked it quite high with the extra 100 percent power offset 400 watts through the card but did something like more than 17 hundred Hertz so that'll be fun content check back for that you can subscribe for it patreon.com slash gamers Nexus if you'd like to help us out directly in this type of reporting or gamers Nexus squarespace.com to pick up a shirt like this one thank you for watching I'll see you all next time you